Prevent LoopVectorizer and SLPVectorizer running if the target has no vector registers.

XCore target: Add XCoreTargetTransformInfo
This is where getNumberOfRegisters() resides, which in turn returns the
number of vector registers (=0).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190936 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/XCore/XCore.h

@@ -31,6 +31,8 @@ namespace llvm {
                                    CodeGenOpt::Level OptLevel);
   ModulePass *createXCoreLowerThreadLocalPass();
 
+  ImmutablePass *createXCoreTargetTransformInfoPass(const XCoreTargetMachine *TM);
+
 } // end namespace llvm;
 
 #endif

lib/Target/XCore/XCoreTargetMachine.cpp

@@ -70,3 +70,11 @@ bool XCorePassConfig::addInstSelector() {
 extern "C" void LLVMInitializeXCoreTarget() {
   RegisterTargetMachine<XCoreTargetMachine> X(TheXCoreTarget);
 }
+
+void XCoreTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+  // Add first the target-independent BasicTTI pass, then our XCore pass. This
+  // allows the XCore pass to delegate to the target independent layer when
+  // appropriate.
+  PM.add(createBasicTargetTransformInfoPass(this));
+  PM.add(createXCoreTargetTransformInfoPass(this));
+}

lib/Target/XCore/XCoreTargetMachine.h

@@ -57,6 +57,8 @@ public:
 
   // Pass Pipeline Configuration
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
+
+  virtual void addAnalysisPasses(PassManagerBase &PM);
 };
 
 } // end namespace llvm

lib/Target/XCore/XCoreTargetTransformInfo.cpp

@@ -0,0 +1,85 @@
+//===-- XCoreTargetTransformInfo.cpp - XCore specific TTI pass ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file implements a TargetTransformInfo analysis pass specific to the
+/// XCore target machine. It uses the target's detailed information to provide
+/// more precise answers to certain TTI queries, while letting the target
+/// independent and default TTI implementations handle the rest.
+///
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "xcoretti"
+#include "XCore.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/CostTable.h"
+using namespace llvm;
+
+// Declare the pass initialization routine locally as target-specific passes
+// don't havve a target-wide initialization entry point, and so we rely on the
+// pass constructor initialization.
+namespace llvm {
+void initializeXCoreTTIPass(PassRegistry &);
+}
+
+namespace {
+
+class XCoreTTI : public ImmutablePass, public TargetTransformInfo {
+  const XCoreTargetMachine *TM;
+
+public:
+  XCoreTTI() : ImmutablePass(ID), TM(0) {
+    llvm_unreachable("This pass cannot be directly constructed");
+  }
+
+  XCoreTTI(const XCoreTargetMachine *TM)
+      : ImmutablePass(ID), TM(TM) {
+    initializeXCoreTTIPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual void initializePass() {
+    pushTTIStack(this);
+  }
+
+  virtual void finalizePass() {
+    popTTIStack();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    TargetTransformInfo::getAnalysisUsage(AU);
+  }
+
+  static char ID;
+
+  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+    if (ID == &TargetTransformInfo::ID)
+      return (TargetTransformInfo*)this;
+    return this;
+  }
+
+  unsigned getNumberOfRegisters(bool Vector) const {
+    if (Vector) {
+       return 0;
+    }
+    return 12;
+  }
+};
+
+} // end anonymous namespace
+
+INITIALIZE_AG_PASS(XCoreTTI, TargetTransformInfo, "xcoretti",
+                   "XCore Target Transform Info", true, true, false)
+char XCoreTTI::ID = 0;
+
+
+ImmutablePass *
+llvm::createXCoreTargetTransformInfoPass(const XCoreTargetMachine *TM) {
+  return new XCoreTTI(TM);
+}

lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -909,6 +909,11 @@ struct LoopVectorize : public LoopPass {
     DT = &getAnalysis<DominatorTree>();
     TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
 
+    // If the target claims to have no vector registers don't attempt
+    // vectorization.
+    if (!TTI->getNumberOfRegisters(true))
+      return false;
+
     if (DL == NULL) {
       DEBUG(dbgs() << "LV: Not vectorizing because of missing data layout");
       return false;

lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -1572,6 +1572,11 @@ struct SLPVectorizer : public FunctionPass {
     StoreRefs.clear();
     bool Changed = false;
 
+    // If the target claims to have no vector registers don't attempt
+    // vectorization.
+    if (!TTI->getNumberOfRegisters(true))
+      return false;
+
     // Must have DataLayout. We can't require it because some tests run w/o
     // triple.
     if (!DL)

test/Transforms/BBVectorize/xcore/no-vector-registers.ll

@@ -0,0 +1,18 @@
+; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -instcombine -gvn -S -mtriple=xcore | FileCheck %s
+
+target datalayout = "e-p:32:32:32-a0:0:32-n32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f16:16:32-f32:32:32-f64:32:32"
+target triple = "xcore"
+
+; Basic depth-3 chain
+define double @test1(double %A1, double %A2, double %B1, double %B2) {
+; CHECK-LABEL: @test1(
+; CHECK-NOT: <2 x double>
+  %X1 = fsub double %A1, %B1
+  %X2 = fsub double %A2, %B2
+  %Y1 = fmul double %X1, %A1
+  %Y2 = fmul double %X2, %A2
+  %Z1 = fadd double %Y1, %B1
+  %Z2 = fadd double %Y2, %B2
+  %R  = fmul double %Z1, %Z2
+  ret double %R
+}

test/Transforms/LoopVectorize/xcore/no-vector-registers.ll

@@ -0,0 +1,23 @@
+; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-unroll=2 -S -mtriple=xcore | FileCheck %s
+
+target datalayout = "e-p:32:32:32-a0:0:32-n32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f16:16:32-f32:32:32-f64:32:32"
+target triple = "xcore"
+; The xcore target has no vector registers, so loop should not be vectorized.
+;CHECK-LABEL: @f(
+;CHECK: entry:
+;CHECK-NOT: vector.body
+;CHECK-NEXT: br label %do.body
+define void @f(i8* nocapture %ptr, i32 %len) {
+entry:
+  br label %do.body
+do.body:
+  %ptr.addr.0 = phi i8* [ %ptr, %entry ], [ %incdec.ptr, %do.body ]
+  %len.addr.0 = phi i32 [ %len, %entry ], [ %dec, %do.body ]
+  %incdec.ptr = getelementptr inbounds i8* %ptr.addr.0, i32 1
+  store i8 0, i8* %ptr.addr.0, align 1
+  %dec = add nsw i32 %len.addr.0, -1
+  %tobool = icmp eq i32 %len.addr.0, 0
+  br i1 %tobool, label %do.end, label %do.body
+do.end:
+  ret void
+}

test/Transforms/SLPVectorizer/xcore/no-vector-registers.ll

@@ -0,0 +1,24 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=xcore  | FileCheck %s
+
+target datalayout = "e-p:32:32:32-a0:0:32-n32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f16:16:32-f32:32:32-f64:32:32"
+target triple = "xcore"
+
+; Simple 3-pair chain with loads and stores
+; CHECK: test1
+; CHECK-NOT: <2 x double>
+define void @test1(double* %a, double* %b, double* %c) {
+entry:
+  %i0 = load double* %a, align 8
+  %i1 = load double* %b, align 8
+  %mul = fmul double %i0, %i1
+  %arrayidx3 = getelementptr inbounds double* %a, i64 1
+  %i3 = load double* %arrayidx3, align 8
+  %arrayidx4 = getelementptr inbounds double* %b, i64 1
+  %i4 = load double* %arrayidx4, align 8
+  %mul5 = fmul double %i3, %i4
+  store double %mul, double* %c, align 8
+  %arrayidx5 = getelementptr inbounds double* %c, i64 1
+  store double %mul5, double* %arrayidx5, align 8
+  ret void
+}
+